In what key is Hey Joe? A Keygram.

---

The three brightest lines are E minor, E major and Eb major. In reality, Hey Joe is in E blues. Because the E blues scale incorporates both major and minor chords. On the link I referenced to, one can see an analysis of why this probably is. In a simplified manner of speaking, the guitar is tuned in E blues, in a way. All of the strings in standard tuning (EADGBE) are part of the E blues scale, the E pentatonic and the E minor scale. It is suggested that Jimi Hendrix wrote Hey Joe being highly influenced by the anatomy or tuning of the guitar itself. This suggests that he probably didn’t write it in his head, he did not hear a melody and immediately knew what notes to play, even though he does have PP, he probably still wrote Hey Joe by trying things out on his guitar. This is fascinating, in a way, since the idea of PP can sound superhuman, and one might expect musicians with PP to immediately know which notes to play the second an idea comes to mind.

The structure of Hideaway by Jacob Collier (tonal and timbral self-similarity matrices)

---

In the pitch graph, you can see very clear boxes, especially around the 200-300 mark. These indicate novelty. The 200-300 mark is the bridge, which is why it is the clearest box (since bridges are, generally, melodically very different from the verse and chorus). You can also see some diagonal lines around the 75-200 mark, which indicate repetition. In the timbre graph however, you do not see these diagonal lines, apart from the one through the middle that indicates that we’re graphing the same song against itself. This means that there seems to be no or very little repetition within timbre. However, since the whole timbre matrix is almost completely lilac, this indicates that there is very little timbral change.

The keys used by musicians with PP vs musicians with RP

---

There are no major differences, but as you might be able to see, the key distribution for the PP playlist are slightly more uniform than that of the RP playlist. The highs are slightly lower for the RP distribution, and the lows are slightly higher. What one can ask is if there truly is a tendency for people with perfect pitch to vary the keys they use more than people with RP.

What about tempo?

---

The PP distribution clearly has a high spike around 100 beats per minute (bpm). The highest spike of RP is at an even lower bpm, although the bulk is higher than that. Difference in the mean will be analyzed in the next page.

Is there an emotional difference between the two playlists? Valence and energy

---

To look at the emotional differences in the PP and the RP playlist, the variables of valence, energy, mode and tempo, as measured by spotify, were used. High valence is considered as emotionally positive, whereas low valence is considerd as emotionally negative. T-tests were used to find any differences in means. Although “mode” is categorical, it was analyzed as a continuous variable for ease, this does not effect the p-value due to the t-test being a general linear model. With an alpha level of 0.05 none of the emotion variables were significant, although valence was very close to being significant.

Two-sided t-tests

Variable	PP Mean	RP Mean	P-value
Tempo	112.9424091	113.3670357	0.8866453
Energy	0.3720677	0.3424691	0.2750405
Valence	0.4379536	0.3908816	0.0563126
Mode	0.6363636	0.622449	0.7699918

From this, one can conclude there probably aren’t any major differences in the musical emotional expression between people with PP and people with RP.

k-nearest Neighbour Classification

---

k-nearest neighbour classification seems to be doing quite well. A large problem is that I had to randomly select only a part of both playlists due to a lack of processing power, which may have resulted in genre-specific biases. In the next tab, we’ll take a look at which aspects predict the differences between the playlists.

What aspects are predictive? Using a Tree Model to indicate feature importance

---

These results are a bit weird. The biggest factor seems to be loudness, followed by c01, which is a timbre feature that also depicts loudness in a way. If we do a t-test for loudness, the results are insignificant. The reason why the aspect of loudness seems to be a good predictor is probably because I had to limit the analysis to 40 tracks per playlist. Since the sampling for these 40 tracks was done randomly, the difference between these two samples are probably just genre differences. I would have liked to have done this with the complete playlists, but unfortunately my laptop could not handle this. That’s why I’ll leave the classification algorithms out of the conclusion.